The objective of the proposed research is to understand how entire populations of retinal neurons convey visual information to the brain, how the activity of the retinal network activity emerges from its elements, and how network function serves the needs of the organism. Until now, progress on these basic problems has been limited by lack of: (a) direct experimental access to the activity of a complete population of neurons, and (b) models of multineuronal responses that allow us to understand how complete populations of neurons interact to represent information. We have recently developed techniques for recording from complete populations of retinal ganglion cells (RGCs) in isolated macaque monkey retina, and approaches to modeling these responses that provide great promise in understanding how the entire network encodes the visual scene. We will combine these powerful new techniques to address the following aims: (1) How do sensory inputs, nonlinearities, noise, and inter-connections combine to determine the detailed spiking patterns in large ensembles of RGCs? (2) How effectively can visual stimuli be decoded based on the ensemble firing patterns of RGCs, and how does stimulus discriminability depend on the fine temporal structure of spike trains? (3) What aspects of the visual stimulus are most effectively encoded by ensemble RGC activity, and to what degree do these reflect the perceptual abilities of humans and the structure of the natural visual environment? Relevance: Because retinal ganglion cells transmit all visual information to the brain, understanding how they collectively encode visual information is a fundamental aspect of understanding vision, in health and in disease. The proposed work will for the first time allow us to explain the spike trains of a complete population of cells in a single framework that incorporates their response properties, sources of physiological noise, and network connectivity. Each of these ultimately contribute to the healthy function of the visual system, while disruption of each will degrade the visual signals transmitted by the RGC population to the brain in ways that may be predicted and understood with the proposed approach. Furthermore, prosthetic devices to replace retinal function, which are now being tested in humans, will eventually need to reproduce the normal patterns of spiking activity in order to provide natural visual signals to the brain. Therefore, our recent experiments using electrical stimulation with multi-electrode arrays for prosthetic design will benefit greatly from the proposed work. In summary, knowing how the entire retinal network encodes the visual scene in spike trains is a key element in understanding the healthy visual system and designing prosthetic treatments for retinas damaged by disease.